01 / interactive lesson
Neural
network.
A pile of simple filters that learns to spot patterns.
001 AI without the fog
AI, only
simpler.
Click, watch and see what really happens under the hood of artificial intelligence. No jargon. No 40-slide lecture. Just the moment: “oh, I get it.”
07ideas
to start
to start
00text
walls
walls
∞clicks
to “aha”
to “aha”
watch the system
not magic / mechanics
INPUT
OUTPUT / 0.87
layerdata
layerspatterns
answercat?
01small networkmany connections
002 The first seven building blocks
Start with the dots.
Then get the aha.
Pick a concept. Open the experiment. Stay as long as you need.
07lessonsNeural
02 / interactive lesson
Gradient
descent.
A tiny compass that helps AI make the next guess better.
03 / interactive lesson
Embed
dings.
Turn words into locations so related ideas can sit close together.
04 / interactive lesson
Atten
tion.
A weighted spotlight that helps a model focus on useful context.
05 / interactive lesson
Feed-
forward.
A private workshop that expands, activates and compresses one token signal.
06 / interactive lesson
Trans
former block.
The repeatable circuit that mixes context, preserves a route and rewrites the signal.
07 / flagship interactive lesson
How it
works.
Follow one sentence from text to the next-token distribution, with every handoff left visible.
003 Machine learning / field guides
Build the idea.
Then build a model.
Careful starting points for people learning machine learning.
10guidesMachine learning
for beginners.
Features, labels, data splits and a practical learning path.
Open the beginner guideBasic project
ideas.
Small datasets, a reproducible example and honest evaluation.
Explore starter projectsCNN layers,
explained.
Follow tensor shapes from pixels to class scores.
Open the CNN guideBuild a network
from scratch.
Use Python and NumPy to follow weights, gradients and an XOR prediction.
Open the Python tutorialRetrieval-
augmented generation.
See how a search step brings relevant source passages into an LLM’s context.
Open the RAG guideLLM
temperature.
Move a toy next-token distribution and see how temperature changes the odds.
Open the temperature guideAI context
windows.
See what fits in one request and how to leave room for a useful answer.
Open the context guideWhat is a token
in AI?
See how a tokenizer turns text into pieces, IDs and vectors, and why tokens are not words.
Open the token guideLLM vs.
generative AI.
See where language models overlap with the broader family of systems that generate content.
Open the comparison guideIndirect prompt
injection.
See how instructions inside external content can redirect an AI tool—and what the example does not prove.
Open the security example003 How we teach
Three moves.
No stress.
Every idea gets one picture, one interaction and one sentence you can carry with you.
01see it
Picture first.
Before a definition, watch the mechanism move.
02touch it
Then interact.
Change one input and see the result move with it.
03connect it
End with aha.
One clear sentence connects the picture to real AI.
004 The big idea
AI is not magic.
It is signal.
Information comes in. Something transforms it. An answer comes out. We will unpack the details one by one.
the human version / no equations∞ This is only the beginning
See you
under the hood.
The next question is already waiting. Pick a field guide, change a variable and keep the mechanism visible.
New practical guide
LLM Hallucination Examples: Spot Fabricated Facts, Citations and Contradictions
See invented examples of false facts, fake citations, contradictions, and overconfident answers, plus a checklist for verifying an LLM response.
New machine-learning field guide
Building a Large Language Model From Scratch: The Real Workflow
Follow the path from a permitted text corpus to tokens, transformer training and meaningful evaluation—and see what a small learning project can realistically do.